Inflatable space station modules, successful gene therapy for aging, and neuromorphic computing.

May 29, 2016

Now that I've got some spare time (read: Leandra's grinding up a few score gigabytes of data), I'd like to write up some stuff that's been floating around in my #blogfodder queue for a couple of weeks.

First up, private-sector aerospace engineering and orbital insertion contractor SpaceX announced not too long ago announced that one of their unmanned Dragon spacecraft delivered an inflatable habitat module to the International Space Station. Following liftoff from Cape Canaveral the craft executed a rendezvous with the ISS in low earth orbit, where the ISS' manipulator arm grappled the craft. In addition to supplies and freight necessary for crew and station one of Bigelow Aerospace's inflatable station modules. For a space station peripheral the deflated BEAM (Bigelow Expandable Activity Module) is remarkably small (1,360 kilographs of mass, 1.7 meters long, 2.4 meters in diameter), but when completely filled with atmosphere it grew to a full size of 3.2 meters in length by 4 meters in diameter (I think I got those matched up). The current gameplan is to slowly but carefully inflate but not use the module to see how it acts in microgravity; remember that this has never been attempted before so science is being done at the same time that history is being made. While this seems overly cautious there are good (albeit not well advertised) reasons for this: The phenomenon of outgassing (note: SSL cert was issued by NASA's CA, so your browser probably doesn't trust it), or materials one would expect to be stable beause they're usually on Earth emitting gases that can leave films on surfaces (or are potentially toxic in vivo) was first observed in early photogrammetry satellites. Thus, the experimental module is instrumented, probably to determine whether or not (and if so, how much) the construction materials will outgas while installed; the results will be used to provide data when Bigelow Aerospace designs the next iteration of the BEAM. Outgassing aside (because that's the phenomenon I have the most experience with) NASA and Bigelow are also interested in tracking how the BEAM stands up overall (it's a semiflexible pressurized envelope in a vacuum so how well the seams and structural members hold up are a major concern), how well it withstands micrometeoroid impacts (impacts with space dust, basically), how much radiation makes it inside the module over time (pretty much the big issue if this style of module will ever be used for habitation, to say nothing of experiments being corrupted), and, of course, whether or not it leaks.

At the end of the twenty-four month experiment, the BEAM will be sealed up, detached from the ISS, and jettisoned with the assistance of the MSS, whereupon its orbit will decay and it will eventually burn up upon re-entry.

Last year I attended a concerence at which Liz Parrish of BioViva USA, Incorporated gave a presentation on using gene therapy to combat the effects of the aging process at the cellular level. She announced while on stage that her company had begun human trials; what came out later was that she herself was the first clinical test subject. Earlier this year (this gives you an idea of how long these links have been kicking around in my collection) BioViva USA announced the results of Liz's experiment: Twenty years of telomere-related aging were successfully reversed. Telomeres, lenths of repeating base pairs of DNA at the ends of chromosomes help protect them from damage or degredation when cells reproduce. Telomeres additionally prevent chromosomes from adhering to one another and control when a particular set of chromosomes has been replicated too many times to be safe, resulting in apoptosis. Prior to the experiment Spectra Laboratories of Houston, TX extracted a sample of Parrish's blood and extracted the T-lymphocytes, one of the few blood cells that has DNA in it (erythrocytes don't) and examined the lengths of the telomeres therein to establish a baseline of approximately 6,710 base pairs. After this phase of the experiment was over they did the same thing and compared the two - lo and behold, the lengths of her telomeres post-experiment were longer, averaging 7,330 base pairs. Her telomeres had grown significantly in just six months. The findings were independently verified by two separate organizations, HEALES and the Biogerontology Research Foundation. This implies that the telomeres of other kinds of cells throughout her body are also showing signs of telomere regeneration. The original experimental plan was to run for one year with monthly samples and analyses thereof so this represents an unexpected and surprising turn of events. Gentlebeings, stand up another set of sensor nets and start scanning for additional announcements from BioViva this year.

An up-and-coming technology in the last couple of years is neuromorphic computing, or processing information with software or circuitry that attempts to replicate the structure and patterns of organic neurons. Two fairly serious advances were announced in the last couple of months and are on the market at this time. The first was a joint announcement by the Lawrence Livermore National Laboratory and IBM and involves something they call the Scale-Up Synaptic Supercomputer (NS16e). The NS16e is comprised of sixteen of IBM's TrueNorth synaptic processing chips arranged in a 4x4 fully connected grid, resulting in an overall architecture of 16 million artificial neurons and 256 million synapses between them, resulting in a hardware pattern recognition system that, as far as I know, hasn't been implemented in such a compact or efficient way before. Each TrueNorth chip consumes just 70mW of power, meaning that the NS16e as a whole (ancilliary circuitry and all) runs on a paltry 2.5 watts of power (in contrast, your average Macbook Pro requires between 43 and 205 watts of power) and costs in the neighborhood of one million US dollars. The NS16e also has a complete development and implementation suite, including a circuit simulator for testing code, its own programming language (no word on what it looks like yet), development kit, sample code, and tools for assembling different kinds of neural networks. The other advance that you'll probably be interested in is the Fathom Neural Compute Stick from Movidius. The Fathom is referred to as a "deep learning accelerator," a hardware coprocessor engineered to run software neural networks in realtime so the CPU doesn't have to so performance will be better. The interesting bit is that the Fathom has the form factor of your average flash drive, so you can plug it into a USB port on one of your boxen and kick up the performance of your deep learning implementation with very little development effort. Reading through the spec sheet for the Fathom shows that it's based around a VPU (Vision Processing Unit) called the Myriad 2, which are extremely low power devices (consuming less than a watt of power when running) and is designed to be very flexible. The Fathom can run anything from deep neural networks for image processing to arbitrary machine learning constructs programmed using Google's open source TensorFlow libary. Oh, and it runs on Linux and has bindings for Python 3.x. No final price has been announced but the figure $100us keeps getting bandied about in the news because Movidius wants as many people as possible to get hold of it, something that's going to give neural nework hacking a shot in the arm because the Fathom is significantly cheaper than the CUDA-enabled graphics cards many of us use with our projects. Interested hackers are going to want to keep a sharp watch for the moment these suckers hit the open market.